How to pass additional arguments to "az ml batch-endpoint invoke"?

ShilpaGopal 100

I am trying to invoke batch endpoint from CLI via azure devops pipeline. I am using the below cli command. I am following the azure batch cli documentation link

JOB_NAME=$(az ml batch-endpoint invoke -n  temperature-endpoint --input-type uri_folder --input azureml://datastores/temperature/paths/SourceData/Batch --output-path azureml://datastores/temperature/paths/Output --query name -o tsv --debug)

I want to pass the model name & model version as parameters. How can I set the model name and model version in the above invoke command? I tried adding the argument : "--set inputs.modelname=temperature_mdl" to the above invoke command. But I get error:

marshmallow.exceptions.ValidationError: {'inputs': ['Unknown field.']}

I tried passing : "--file inputs.yml" to the above invoke command. The contents of inputs.yml is as below:

inputs:
  batchinputdatafolder:
    type: uri_folder
    path: azureml://datastores/temperature/paths/SourceData/Batch
  modelname:
    type: string
    default: temperature_mdl
  modelversion:
    type: string
    default: 1
outputs: 
  batchoutputdatafolder:
    type: uri_folder
    path: azureml://datastores/temperature/paths/Output

But I get the error below error even through the inputs.yml file is in the same directly from where I am executing invoke command.

ERROR: cli.azure.cli.core.azclierror: No such file or directory: inputs.yml

Question: How do I pass additional arguments (other than data path) in az ml batch-endpoint invoke CLI?

romungi-MSFT 42,206 Reputation points Microsoft Employee

2024-01-15T05:39:26.6+00:00

@ShilpaGopal I think every azure ML endpoint provides a simple interface for creating and managing model deployments. Each endpoint can have one or more deployments. Each deployment can have a specific model and its version deployed. So, I think you should first create a specific model deployment with a model and its version. For example, See this example to create a specific model deployment. You can then use the deployment name in the batch-endpoint invoke command with --deployment-name parameter. The GH samples also provide this scenario to create a deployment and an endpoint. You can use multiple deployments under an endpoint and invoke a specific model with --deployment-name parameter. Thanks!!
ShilpaGopal 100 Reputation points

2024-01-15T08:29:43.6233333+00:00

@romungi-MSFT

I have created a batch rest endpoint with a default deployment. The deployment points to a component rather than a model.

I have followed the below code: https://github.com/Azure/azureml-examples/tree/main/cli/endpoints/batch/deploy-pipelines/batch-scoring-with-preprocessing

The component is another pipeline as below:

I am trying to invoke the batch default deployment using command: (I have replaced the endpoint name with a dummy one)

JOB_NAME=$(az ml batch-endpoint invoke -n temperature-endpoint --input-type uri_folder --input azureml://datastores/temperature/paths/SourceData/Batch --output-path azureml://datastores/temperature/paths/Output --query name -o tsv --debug)

To the above invoke command I also want to pass 2 custom parameters model_name & model_version. The Azure documentation https://learn.microsoft.com/en-us/cli/azure/ml/batch-endpoint?view=azure-cli-latest#az-ml-batch-endpoint-invoke has a parameter called --set. How can I use the "--set" parameter to pass model_name & model_version in invoke command?
ShilpaGopal 100 Reputation points

2024-01-15T08:35:09.3233333+00:00

@romungi-MSFT
romungi-MSFT 42,206 Reputation points Microsoft Employee

2024-01-17T09:17:00.71+00:00

@ShilpaGopal For this example, based on the screenshot, it could be: --set inputs.modelname.path=test
G Cocci 211 Reputation points Microsoft Employee

2024-04-18T13:00:22.64+00:00

@ShilpaGopal I have the same problem as you, were you able to solve it?

Thanks!